尚未偵測到更新。GitHub 發布版本後會顯示在這裡。
依最近已完成的 8 小時觀測回合間的 GitHub 星號增量排序
個人 GitHub 帳號下公開儲存庫的擁有者可以登入並驗證所有權，自行管理介紹，無需管理員核准。
登入並產生驗證檔案，在 30 分鐘內將其提交至此儲存庫的預設分支。其他分支或分叉儲存庫中的檔案無效。然後在下方檢查檔案，即可無需管理員核准直接管理專案。請保留檔案以維持管理權限。
如需登入，將前往登入頁面。登入後請回到此處，再次點擊所選的驗證按鈕。
在專案資料夾中執行 /nomorevibe verify，即可驗證所有權並自行管理介紹。
一分鐘完成驗證Jag's AI Map: an interactive map of what's heating up in AI (English sources, optional local LLM entity extraction)
翻譯準備中
取自儲存庫 README
正在載入完整 README。
檢視完整 README ↗An interactive map of what's heating up in AI over the last 7 days: companies, models, products, open-source projects, people, hardware and topics, sized and shaded by how much attention they get. Inspired by AI MAP by carlwow, built independently with English sources.
./run.sh # fetch news, extract entities, score heat -> web/data/data.json (1-5 min)
./serve.sh # open http://localhost:8765
./run.sh --use-cache # rebuild from the last fetch (no network), e.g. after tweaking scoring
Needs only Python 3 (standard library). web/ is a static site: D3 is vendored in web/vendor/.
pipeline/sources.py): Hacker News (>30 points), TechCrunch, The Verge, VentureBeat,
MIT Technology Review, Ars Technica, Wired, The Decoder, Simon Willison, OpenAI, Google AI, Google
DeepMind, NVIDIA, Hugging Face and AWS ML blogs, Hugging Face daily papers and trending models, new
AI repos on GitHub. Off-topic Hacker News stories are dropped.pipeline/entities.py): a curated dictionary of ~200 AI entities, automatic discovery of
versioned model names (GPT-6.1, Claude Opus 5.5, Qwen3.8...), trending Hugging Face models and new
GitHub repos, plus optional LLM extraction.run.sh reads it from
.env or, if present, the strata provider in a pi config (~/.pi/agent/models.json). Override with AIMAP_LLM_BASE_URL, AIMAP_LLM_MODEL,
AIMAP_LLM_API_KEY, or disable with AIMAP_LLM=off.web/data/data.json: 7 days, 24 h half-life, trend vs 24 h ago) and
Today (web/data/today.json: 24 h, 6 h half-life, trend vs 6 h ago), switched in the page header.pipeline/build.py): each story adds source weight x popularity boost x 0.5^(age/24h)
to every entity it mentions. "New" = not in the dictionary and first seen this week; up arrow =
heat at least 1.5x its value 24 h earlier. Links = entities mentioned in the same stories.
data/history.json keeps first-seen dates and a heat series for the sparklines.| Path | What |
|---|---|
pipeline/ | Fetch, entity recognition, scoring (Python, stdlib only) |
web/ | The site: index.html, app.js (D3 graph), style.css, data/data.json |
data/ | history.json (kept), stories-cache.json (last fetch, git-ignored) |
.github/workflows/update.yml | 12-hourly refresh + GitHub Pages deploy (for public hosting) |
deploy/local.ai-map.refresh.plist | Optional 12-hourly refresh on this Mac (launchd) |
logs/ | Build logs |
MIT licence (see LICENSE). D3 is ISC-licensed (web/vendor/D3-LICENSE.txt). Headlines link to their publishers.
https://ai-news.jagular.io, hosted on a home Linux server (Ubuntu) that also runs the Qwen model.
| Piece | Where | Notes |
|---|---|---|
| Code + data | /opt/ai-map | Same tree as here. .env (root-only) points the LLM step at http://127.0.0.1:8080/v1 (strata, Qwen3.8-Flash) |
| Hourly refresh | ai-map-refresh.timer → ai-map-refresh.service runs /opt/ai-map/run.sh | systemctl list-timers ai-map-refresh.timer; logs: journalctl -u ai-map-refresh and /opt/ai-map/logs/ |
| Web server | nginx site /etc/nginx/sites-available/ai-news | Static files from /opt/ai-map/web, alongside the existing emailbrain/trials sites |
| Public access | Cloudflare Tunnel jag-ai-map (/etc/cloudflared/config.yml, service cloudflared) | Only ai-news.jagular.io is routed; anything else returns 404. DNS CNAME created by cloudflared tunnel route dns |
Copies of these config files (no secrets) are in deploy/server/.
Qwen answers are cached per story in data/llm-cache.json, so hourly runs only send new headlines.
Update the server after changing code here (keeps the server's .env, data and caches):
rsync -a --exclude logs/ --exclude screenshots/ --exclude data/ --exclude .env --exclude web/data/ ./ root@<your-server>:/opt/ai-map/
Create .env next to run.sh (never committed):
AIMAP_LLM_BASE_URL=http://127.0.0.1:8080/v1 # any OpenAI-compatible server (vLLM, llama.cpp, Ollama...)
AIMAP_LLM_MODEL=your-model-id
AIMAP_LLM_API_KEY=your-key
Without it, the map still builds using the dictionary and model-name discovery.